CN109462520A - Network flow resource Tendency Prediction method based on LSTM model - Google Patents
Network flow resource Tendency Prediction method based on LSTM model Download PDFInfo
- Publication number
- CN109462520A CN109462520A CN201811375055.5A CN201811375055A CN109462520A CN 109462520 A CN109462520 A CN 109462520A CN 201811375055 A CN201811375055 A CN 201811375055A CN 109462520 A CN109462520 A CN 109462520A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- lstm
- training
- flow
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000005070 sampling Methods 0.000 claims description 14
- 238000004891 communication Methods 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 4
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 210000002364 input neuron Anatomy 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 210000005036 nerve Anatomy 0.000 claims 1
- 230000006870 function Effects 0.000 abstract description 3
- 230000006399 behavior Effects 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000008901 benefit Effects 0.000 abstract description 2
- 230000008878 coupling Effects 0.000 abstract description 2
- 238000010168 coupling process Methods 0.000 abstract description 2
- 238000005859 coupling reaction Methods 0.000 abstract description 2
- 230000007774 longterm Effects 0.000 abstract description 2
- 230000000306 recurrent effect Effects 0.000 abstract description 2
- 230000006403 short-term memory Effects 0.000 abstract description 2
- 239000000523 sample Substances 0.000 description 23
- 230000000694 effects Effects 0.000 description 15
- 210000002569 neuron Anatomy 0.000 description 10
- 235000019580 granularity Nutrition 0.000 description 7
- 230000004907 flux Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- ZEFNOZRLAWVAQF-UHFFFAOYSA-N Dinitolmide Chemical compound CC1=C(C(N)=O)C=C([N+]([O-])=O)C=C1[N+]([O-])=O ZEFNOZRLAWVAQF-UHFFFAOYSA-N 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- BULVZWIRKLYCBC-UHFFFAOYSA-N phorate Chemical compound CCOP(=S)(OCC)SCSCC BULVZWIRKLYCBC-UHFFFAOYSA-N 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000000714 time series forecasting Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
- H04L43/0876—Network utilisation, e.g. volume of load or congestion level
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Environmental & Geological Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention belongs to network flow resource Situation Forecast Technique field, especially a kind of network flow resource Tendency Prediction method based on LSTM model.It is an object of the invention to propose that the shot and long term suitable for flow space-time non-linear behavior remembers (Long Short-Term Memory, LSTM) Recognition with Recurrent Neural Network prediction model.The characteristics of models coupling actual flow data burst is relatively strong and Long-range dependence progress model training and quantitative forecast.Beneficial effects of the present invention are that the present invention effectively raises forecasting accuracy to the forecast function of network flow by LSTM model realization, and when data sample otherness is larger, the advantage of LSTM prediction model is more significant.
Description
Technical field
The invention belongs to network flow resource Situation Forecast Technique field, especially a kind of network flows based on LSTM model
Measure resource Tendency Prediction method.
Background technique
As universal and development, the network management of network face many challenges.Wireless sense network, Ad hoc network,
Space-based network etc. lengthy and jumbled heterogeneous network equipment makes topological structure increasingly sophisticated, and frequent information exchange makes
Network flow is increased sharply, and the complexity and uncertainty description ability of network operation situation accordingly reduce.Volume forecasting has become stream
Measure engineering, the key problem of congestion control and network O&M management study.
Due to communication network multi-protocols characteristic and service source it is sudden, network load be often maintained at big flow burst with
The state being alternately present opposing stationary for a long time, so that the non-linear behavior of data on flows is particularly evident.Therefore, adaptation is probed into
In the Predicting Technique of the more variability features of communication network traffic data, the flow value of dynamic prediction future time instance will predict error
Control in a certain range, preferably can provide data supporting for communication network information feedback and scheduling of resource, this is to communication
Network is disposed for a long time has important directive significance.
Volume forecasting is referred to by carrying out sample to the suitable model of the historical data selection after data prediction
It practises, achievees the effect that predict the same index value in the following a certain section of moment.In order to improve the accuracy of prediction, selected in model
Select the characteristics of being adapted to historical time sequence.In recent years, abundant with machine learning and deep learning knowledge background
With the extension of research field, the processing of data is no longer limited to the adaptability of traditional single mathematical model.Based on artificial neural network
The deep learning model of network can be by the adaptive learning of sample, and iteration updates network weight between layers and biasing
Parameter, preferable fit non-linear data handle problem.By abundant training sample and increase hidden layer quantity and hidden neuron
Number is capable of the fitness of larger range adjustment model, has outstanding performance in Forecast of Nonlinear Time Series, but simultaneously because
To the sensibility and dependence of data characteristics, there are biggish othernesses for the model adapted under different scenes.
Communication network traffic can due to the difference of the factors such as network environment, flow collection time span and time scale table
Reveal different characteristics, predicting network flow is difficult to find a kind of general modeling method, and main cause has: 1) data itself
There are stronger randomness, it is not necessarily to certain set rule and occurs, prediction error is difficult to pass through mathematical statistical model
Effective control；2) there are more emergency cases for network flow, lead to the non-linear sharp increase of flow, even if using artificial neuron
The nonlinear models such as network are also difficult to carry out accurate modeling to its changing rule.
Summary of the invention
Goal of the invention of the invention is: in view of the above problems, having probed into and has been suitable for the non-linear spy of flow space-time
The shot and long term of point remembers (Long Short-Term Memory, LSTM) Recognition with Recurrent Neural Network prediction model.The models coupling
The characteristics of actual flow data burst is relatively strong and Long-range dependence progress model training and quantitative forecast.
The technical solution adopted by the present invention is as follows, the network flow resource Tendency Prediction method based on LSTM model, special
Sign is, comprising the following steps:
S1, data prediction: by the resulting communication network OD of fixed frequency down-samplingijFlow data is handled, and is obtained
Time series X (t)=[x under to N number of continuous cycles1,x2,...,xN]；Wherein ODijThe definition of stream is host i to host j's
End-to-end flux；
S2, it establishes LSTM model: determining in neural network input neuron number R, output neuron number M and hidden
Layer neuronal quantity, is completed from input space URTo output space UMMapping, i.e., according to preceding R moment flow value prediction after M
A moment flow value, setting step-length τ carry out phase space decomposition and reconstruction, establish training sample and input the mapping of X to output Y:
Wherein yi=x(m-i)τ；
Wherein, the corresponding trained input sample of every a line in X, length τ are mapped by LSTM model forward-propagating
Obtain output sample label value yi', with true flow rate value yiCarry out error costing bio disturbance, feedback training LSTM model；
S3, according to S1 obtain data, to S2 establish LSTM model into training, obtain trained LSTM model, have
Body are as follows:
S31, training data n is cut into from the data obtained in S1, and by step-length τ is set in S2, by training data point
For+1 sample set of n/ τ, Space Reconstruction is carried out, is stored as input matrix X and its homography Y by column；
S32, matrix is normalized by column, data is transformed into [0,1] range, standardized data is established；
The normalization uses formulaWherein xminAnd xmaxThe minimum value and maximum value of respectively this group sample, xi
To correspond to each flow value in sample group, i takes 1,2 ... τ；
S33, the random initial weight of setting initialize LSTM network, and input sample collection is trained network,
The neural network of 1 output of τ input is established according to parameter corresponding to mean square error PMSE optimal in training；
S4, prediction is obtained using LSTM model trained in step S3 using the method acquisition data on flows of step S1
As a result.
Beneficial effects of the present invention are that the present invention passes through LSTM model realization to the forecast function of network flow, effectively
Forecasting accuracy is improved, and when data sample otherness is larger, the advantage of LSTM prediction model is more significant.
Detailed description of the invention
Fig. 1 is the volume forecasting algorithm flow based on LSTM；
Fig. 2 is experimental data figure, wherein figure (a) is the single OD stream under different sampling granularities, schemes (b) for different OD stream
Flow form；
Fig. 3 is the predicting network flow based on LSTM；
Fig. 4 is the volume forecasting Contrast on effect under different step-lengths and hidden neuron quantity, wherein figure (a) is fixed mind
The influence that different step parameters generate is probed into through first quantity, figure (b) is the shadow that fixed step size probes into different hidden neuron quantity
It rings；
Fig. 5 is the influence under different sampling granularities to flow prediction result；
Fig. 6 is by the prediction case comparing result probing into algorithm under different flow form Yu commonly use other algorithms, wherein right
It include PSO-LSSVM, BP, Elman algorithm than algorithm, wherein figure (a) is training effect comparison diagram, and figure (b) is that prediction is opposite
Error comparison diagram.
Specific embodiment
With reference to the accompanying drawings and examples, the technical schemes of the invention are described in detail.
As shown in Figure 1, being the volume forecasting algorithm flow of the invention based on LSTM, specifically include that
Sample data is obtained according to historical data, i.e., to data prediction: by resulting logical to fixed frequency down-sampling
Communication network ODijStream (end-to-end flux of the host i to host j) data are handled, and the time sequence under N number of continuous cycles is obtained
Arrange X (t)=[x1,x2,...,xN]。
1) by double sampling by the further discretization of data is obtained different sampling weeks for single OD flow data
Time series data under phase；
2) the OD stream for choosing a plurality of different characteristics under consolidated network carries out constant duration sampling, obtains a plurality of time
Sequence data.
Network model design, is designed LSTM network model, determines input neuron number R in neural network, defeated
Neuron number M and hidden neuron quantity out are completed from input space URTo output space UMMapping.It is single for having
The time series forecasting of one quantizating index, according to preceding R moment flow value prediction after M moment flow value can choose step-length τ into
Row phase space decomposition and reconstruction, is converted into the mapping problems of X to Y:
Wherein yi=x(m-i)τ；
Wherein the corresponding trained input sample of every a line in X, length τ are mapped by LSTM model forward-propagating
Obtain output sample label value yi', with true flow rate value yiCarry out error costing bio disturbance, feedback training LSTM model.
The parameter setting of neural network model, such as the logic gate weighting parameter of LSTM network.During prediction, by repeatedly
These parameters are adaptively adjusted for the method for optimizing, the selection criteria of parameter is using the optimal mean square error before terminating algebra K
PMSE, i.e., when reaching convergence, deconditioning and parameter update.
Wherein k indicates current iteration number；
LSTM can be good at the Nonlinear Mapping between learning data, during LSTM model training, each LSTM mind
Input through first cell all includes location mode C of upper momentt-1H is exported with upper moment LSTMt-1And current time inputs XtThree
Part forms, while input gate, forgetting door and the out gate in neural unit complete the function of information selection and information conversion
Can, it is specific as follows:
Forgeing door is used to select the information at a moment to be retained or forget, 1 indicates to retain, and 0 indicates to abandon, wf、bfPoint
Not Wei refreshing input gate correspond to weight and offset parameter, ht-1Indicate last moment output information, XtFor the input at current time:
ft=σ (wf·[ht-1,Xt]+bf)∈{0,1}
Input gate is responsible for calculating reservation information i currently enteredtWith new state information Ct' generation, wi、bi、wc、bcTo lose
Forget corresponding weight and an offset parameter:
So the hidden neuron state of current LSTM is updated to Ct=ft*Ct-1+it*Ct'；
The output information h at out gate expression current timet, it is that last moment state, current input and implicit layer state are total
The result of same-action:
Wherein, wo、boWeight and offset parameter are corresponded to for out gate, with wf、bf、wi、bi、wc、bcTogether participate in training and
Iteration updates deconditioning when convergent iterations number reaches maximum critical condition K, saves current optimized parameter building LSTM
Network model.
Historical data is divided into training set and test set, participates in network training for the data of training set as network inputs,
Parameter update is carried out according to gradient descent algorithm, establishes neural network model, step refinement are as follows:
1) time series data of certain OD stream composition is cut into training data and prediction data, training set is for training
Just
The LSTM network model of beginning, forecast set are used to verify the validity of model after training；
2) step-length τ (value of subsequent time is predicted according to how many historical datas) is set, and by training data according to selected
Step-length is divided into+1 sample set of n/ τ, carries out Space Reconstruction, is stored as input matrix X and its homography Y by column；
3) matrix is normalized by column, data is transformed into [0,1] range, standardized data is established；
Normalization can be effectively reduced the otherness between data, and sample fluctuation can be reduced by being normalized by sample column
Property, make in training process neural network be easier to restrain.Normalization uses formulaWherein xminAnd xmax
The minimum value and maximum value of respectively this group sample, xiTo correspond to each flow value in sample group, i takes 1,2 ... τ；
4) it sets random initial weight to initialize LSTM network, and input sample collection is trained network, root
The neural network of 1 output of τ input is established according to parameter corresponding to mean square error PMSE optimal in training；
5) identical sample is made to prediction data to divide, obtained by trained LSTM network by output valve y 'iWith reality
Value yiIt is compared, the accuracy of model is measured with relative error.
Embodiment
In order to examine flux prediction model effect, this example is tested using Abilene network flow data.Abilene
Network is American education scientific research net, and core network topology includes 12 nodes and 15 two-way links, to source destination node
The service traffics transmitted between (OD to) are primary every sampling in 5 minutes.During experiment has chosen 2003/05/01~2003/05/30
Live network data on flows, total 12*24*7*4=8064 traffic matrix.
S1, data on flows is pre-processed first.It extracts from collected 8064 traffic matrixs from any two
All moment flow value makeup time sequences between node, selected part time series are predicted.
Since the thickness of time granularity may have an impact prediction result, to extracted time series do into
The sampling of one step, and polymerize and obtain the true flow rate value of more coarseness (generally also more stable), Fig. 1 is each grain by taking OD123 as an example
Real traffic is spent, when polymerizeing for fine granularity 5 minutes to 1 hour scale, using the sampling technique for taking peak value[17], and for
1 hour to 1 day at the time of polymerize, choose flow value representative of the mean value as half a day (12 hours).
S2, OD123 volume forecasting is carried out using LSTM network model, training mean square error is as shown in figure 3, with iteration
The increase of number, right value update can converge to optimal value in gradient descent procedures.
S3, predicting network flow:
Adaptive adjusting algorithm is used to make network model during the test every time with best initial weights wf、wi、wc、woWith
Offset parameter bf、bi、bc、boForward-propagating is carried out, the prediction output of LSTM network and relative error are as shown in Figure 3.From figure
As can be seen that relative error SSE is controlled substantially in 20% range.
S4, hyper parameter impact factor are probed into.The hyper parameter tuple (hidden neuron quantity, step-length) of LSTM, i.e.,
[hidden_numbers, eph] there are different degrees of influences to trained and prediction effect.
Under different hidden neuron quantity, LSTM step-length is different to predicted impact, even if in identical hidden neuron
Under several, influence of the step-length to prediction is also different, respectively as shown in Fig. 4 (a) and Fig. 4 (b).By many experiments, in super ginseng
When number [hidden_numbers=5, eph=4], error is relatively small and variation tends towards stability；Reach in training mean square error
When minimum value, Relative Error is also corresponding smaller, illustrates that selecting to train mean square error minimum value is to promote volume forecasting precision
Key.
Model fitness under S5, different time granularity is probed into.Fig. 5 is using LSTM to (5 points under 3 kinds of different sampling rates
Clock, 1 hour, 12 hours) OD123 data on flows carried out trained comparison.From training mean square error curve, sampling
The increase for being partitioned into multiple does not reduce the self-similarity of flow.Under the premise of taking identical training sample (56 groups), training
Mean square error minimum value is held in 0.003 or so, while also illustrating that LSTM network is weaker for data sensitive, for not
There is preferable fitness with the network flow under sampling granularity.
S6, algorithm evaluation.BP networks, Elman network, particle group optimizing SVM (PSO-SVM) and set forth herein
LSTM network have certain prediction effect to flow.In order to compare the prediction effect and generalization ability of above 4 kinds of models,
Multiple data on flows samples with different characteristics are chosen during prediction, OD123, OD109, OD49, OD86 are respective
Flow feature is as shown in Figure 1, here to train mean square error minimum value to carry out parameter optimization for convergence target, for above four
Kind flow, chooses 9 groups of forecast samples respectively, carries out performance comparison using Relative Error average value as evaluation index.
From the point of view of Fig. 6 (a), the convergence effect of LSTM and PSO-SVM and Elman are substantially better than BP neural network, but
PSO-SVM is mutated biggish stream (such as OD49) for data and more sparse data (such as OD109) training error obviously increases
Greatly, and Elman and LSTM is preferable to every kind of data flow convergence effect；Find out from Fig. 6 (b), LSTM network and PSO-SVM
Prediction effect is substantially better than BP network and Elman network, and it is better than neural network that PSO-SVM prediction effect shows mostly, but at certain
Prediction effect deviation is larger on a little OD streams (such as OD49)；From the model training time each in table 1, with other three kinds of models
It compares, PSO-SVM training time longest.
The 1 model training time (s) of table
LSTM network is weaker compared with other models for the OD stream dependence of different data feature, training error and prediction effect
Also relatively preferably, faster training speed is also able to satisfy the demand of the prediction of the dynamic regression in certain accuracy rating.It is surveyed from model
From the point of view of on test result, LSTM solve the problems, such as sudden stronger network flow data it is difficult to predict.Compared to BP,
The models such as Elman and PSO-SVM, LSTM have preferably the network flow data training effect and prediction effect of non-stationary
Performance, have lower dependence simultaneously for data, in different time granularity and different business fluxion it is predicted that on have compared with
Good fitness.
Claims (2)
1. the network flow resource Tendency Prediction method based on LSTM model, which comprises the following steps:
S1, data prediction: by the resulting communication network OD of fixed frequency down-samplingijFlow data is handled, and is obtained N number of
Time series X (t)=[x under continuous cycles1,x2,...,xN]；Wherein ODijThe definition of stream is host i to the end-to-end of host j
Flow；
S2, it establishes LSTM model: determining input neuron number R, output neuron number M and hidden layer nerve in neural network
First quantity is completed from input space URTo output space UMMapping, i.e., flowed according to M moment after the prediction of preceding R moment flow value
Magnitude, setting step-length τ carry out phase space decomposition and reconstruction, establish training sample and input the mapping of X to output Y:
Wherein yi=x(m-i)τ；
Wherein, the corresponding trained input sample of every a line in X, length τ map to obtain by LSTM model forward-propagating
Export sample label value yi', with true flow rate value yiCarry out error costing bio disturbance, feedback training LSTM model；
S3, according to S1 obtain data, to S2 establish LSTM model into training, obtain trained LSTM model, specifically:
S31, it is cut into training data n from the data obtained in S1, and by step-length τ is set in S2, training data is divided into n/ τ
+ 1 sample set carries out Space Reconstruction, is stored as input matrix X and its homography Y by column；
S32, matrix is normalized by column, data is transformed into [0,1] range, standardized data is established；It is described
Normalization uses formulaWherein xminAnd xmaxThe minimum value and maximum value of respectively this group sample, xiFor sample
Each flow value is corresponded in this group, i takes 1,2 ... τ；
S33, the random initial weight of setting initialize LSTM network, and input sample collection is trained network, according to
Parameter corresponding to optimal mean square error PMSE establishes the neural network of 1 output of τ input in training；
S4, data on flows is obtained using the method for step S1, using LSTM model trained in step S3, obtains prediction knot
Fruit.
2. the network flow resource Tendency Prediction method according to claim 1 based on LSTM model, which is characterized in that institute
It states in step S33, the acquisition methods of parameter corresponding to optimal mean square error PMSE are:
In the training process, the parameter of LSTM model is adaptively adjusted by the method for iteration optimizing, the selection criteria of parameter is adopted
With the optimal mean square error PMSE before termination algebra K, i.e., when reaching convergence, deconditioning and parameter update:
Wherein k indicates current iteration number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375055.5A CN109462520B (en) | 2018-11-19 | 2018-11-19 | Network traffic resource situation prediction method based on LSTM model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375055.5A CN109462520B (en) | 2018-11-19 | 2018-11-19 | Network traffic resource situation prediction method based on LSTM model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109462520A true CN109462520A (en) | 2019-03-12 |
| CN109462520B CN109462520B (en) | 2021-12-10 |
Family
ID=65610832
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811375055.5A Expired - Fee Related CN109462520B (en) | 2018-11-19 | 2018-11-19 | Network traffic resource situation prediction method based on LSTM model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109462520B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109921950A (en) * | 2019-03-28 | 2019-06-21 | 北京奇艺世纪科技有限公司 | A kind of network bandwidth requirements prediction technique, device and electronic equipment |
| CN110381524A (en) * | 2019-07-15 | 2019-10-25 | 安徽理工大学 | The mobile flow on-line prediction method of large scene based on Bi-LSTM, system and storage medium |
| CN110445646A (en) * | 2019-07-30 | 2019-11-12 | 广东工业大学 | A kind of method for obligating resource based on attention mechanism and flux prediction model |
| CN110545208A (en) * | 2019-09-23 | 2019-12-06 | 电子科技大学 | Network traffic prediction method based on LSTM |
| CN111277434A (en) * | 2020-01-16 | 2020-06-12 | 北京工业大学 | Network flow multi-step prediction method based on VMD and LSTM |
| CN111382838A (en) * | 2020-01-15 | 2020-07-07 | 南宁市勘察测绘地理信息院 | Urban liquid level elevation waterlogging prediction method, device and equipment |
| CN111585948A (en) * | 2020-03-18 | 2020-08-25 | 宁波送变电建设有限公司永耀科技分公司 | Intelligent network security situation prediction method based on power grid big data |
| CN111783953A (en) * | 2020-06-30 | 2020-10-16 | 重庆大学 | 24-point power load value 7-day prediction method based on optimized LSTM network |
| CN112165402A (en) * | 2020-09-28 | 2021-01-01 | 北京环境特性研究所 | Method and device for predicting network security situation |
| CN112291807A (en) * | 2020-10-15 | 2021-01-29 | 山东科技大学 | Wireless cellular network traffic prediction method based on deep migration learning and cross-domain data fusion |
| CN112330003A (en) * | 2020-10-27 | 2021-02-05 | 电子科技大学 | Periodic capacity data prediction method, system and storage medium based on bidirectional cyclic neural network |
| CN112461537A (en) * | 2020-10-16 | 2021-03-09 | 浙江工业大学 | Wind power gear box state monitoring method based on long-time neural network and automatic coding machine |
| CN113316163A (en) * | 2021-06-18 | 2021-08-27 | 东南大学 | Long-term network traffic prediction method based on deep learning |
| CN114202065A (en) * | 2022-02-17 | 2022-03-18 | 之江实验室 | Stream data prediction method and device based on incremental evolution LSTM |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106845371A (en) * | 2016-12-31 | 2017-06-13 | 中国科学技术大学 | A kind of city road network automotive emission remote sensing monitoring system |
| CN107544904A (en) * | 2017-08-21 | 2018-01-05 | 哈尔滨工程大学 | A kind of prediction model of software reliability based on depth CG LSTM neutral nets |
| CN108062561A (en) * | 2017-12-05 | 2018-05-22 | 华南理工大学 | A kind of short time data stream Forecasting Methodology based on long memory network model in short-term |
-
2018
- 2018-11-19 CN CN201811375055.5A patent/CN109462520B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106845371A (en) * | 2016-12-31 | 2017-06-13 | 中国科学技术大学 | A kind of city road network automotive emission remote sensing monitoring system |
| CN107544904A (en) * | 2017-08-21 | 2018-01-05 | 哈尔滨工程大学 | A kind of prediction model of software reliability based on depth CG LSTM neutral nets |
| CN108062561A (en) * | 2017-12-05 | 2018-05-22 | 华南理工大学 | A kind of short time data stream Forecasting Methodology based on long memory network model in short-term |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109921950A (en) * | 2019-03-28 | 2019-06-21 | 北京奇艺世纪科技有限公司 | A kind of network bandwidth requirements prediction technique, device and electronic equipment |
| CN109921950B (en) * | 2019-03-28 | 2021-08-06 | 北京奇艺世纪科技有限公司 | Network bandwidth demand prediction method and device and electronic equipment |
| CN110381524A (en) * | 2019-07-15 | 2019-10-25 | 安徽理工大学 | The mobile flow on-line prediction method of large scene based on Bi-LSTM, system and storage medium |
| CN110381524B (en) * | 2019-07-15 | 2022-12-20 | 安徽理工大学 | Bi-LSTM-based large scene mobile flow online prediction method, system and storage medium |
| CN110445646A (en) * | 2019-07-30 | 2019-11-12 | 广东工业大学 | A kind of method for obligating resource based on attention mechanism and flux prediction model |
| CN110445646B (en) * | 2019-07-30 | 2022-07-08 | 广东工业大学 | Resource reservation method based on attention mechanism and flow prediction model |
| CN110545208A (en) * | 2019-09-23 | 2019-12-06 | 电子科技大学 | Network traffic prediction method based on LSTM |
| CN111382838A (en) * | 2020-01-15 | 2020-07-07 | 南宁市勘察测绘地理信息院 | Urban liquid level elevation waterlogging prediction method, device and equipment |
| CN111277434A (en) * | 2020-01-16 | 2020-06-12 | 北京工业大学 | Network flow multi-step prediction method based on VMD and LSTM |
| CN111585948A (en) * | 2020-03-18 | 2020-08-25 | 宁波送变电建设有限公司永耀科技分公司 | Intelligent network security situation prediction method based on power grid big data |
| CN111585948B (en) * | 2020-03-18 | 2022-07-26 | 宁波送变电建设有限公司永耀科技分公司 | Intelligent network security situation prediction method based on power grid big data |
| CN111783953A (en) * | 2020-06-30 | 2020-10-16 | 重庆大学 | 24-point power load value 7-day prediction method based on optimized LSTM network |
| CN111783953B (en) * | 2020-06-30 | 2023-08-11 | 重庆大学 | 24-point power load value 7-day prediction method based on optimized LSTM network |
| CN112165402A (en) * | 2020-09-28 | 2021-01-01 | 北京环境特性研究所 | Method and device for predicting network security situation |
| CN112291807A (en) * | 2020-10-15 | 2021-01-29 | 山东科技大学 | Wireless cellular network traffic prediction method based on deep migration learning and cross-domain data fusion |
| CN112291807B (en) * | 2020-10-15 | 2022-09-23 | 山东科技大学 | Wireless cellular network traffic prediction method based on deep migration learning and cross-domain data fusion |
| CN112461537A (en) * | 2020-10-16 | 2021-03-09 | 浙江工业大学 | Wind power gear box state monitoring method based on long-time neural network and automatic coding machine |
| CN112330003A (en) * | 2020-10-27 | 2021-02-05 | 电子科技大学 | Periodic capacity data prediction method, system and storage medium based on bidirectional cyclic neural network |
| CN113316163B (en) * | 2021-06-18 | 2022-04-15 | 东南大学 | Long-term network traffic prediction method based on deep learning |
| CN113316163A (en) * | 2021-06-18 | 2021-08-27 | 东南大学 | Long-term network traffic prediction method based on deep learning |
| CN114202065A (en) * | 2022-02-17 | 2022-03-18 | 之江实验室 | Stream data prediction method and device based on incremental evolution LSTM |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109462520B (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109462520A (en) | Network flow resource Tendency Prediction method based on LSTM model | |
| CN111369042B (en) | Wireless service flow prediction method based on weighted federal learning | |
| CN110223517A (en) | Short-term traffic flow forecast method based on temporal correlation | |
| CN110765700A (en) | Ultrahigh voltage transmission line loss prediction method based on quantum ant colony optimization RBF network | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| CN108446794A (en) | One kind being based on multiple convolutional neural networks combination framework deep learning prediction techniques | |
| CN110470259A (en) | Landslide displacement dynamic prediction method based on LSTM | |
| CN108549770A (en) | The adaptive inversion method of Parameters for Rockfill Dams based on QGA-MMRVM | |
| CN111310965A (en) | Aircraft track prediction method based on LSTM network | |
| CN109063939A (en) | A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network | |
| CN105160441B (en) | It is transfinited the real-time electric power load forecasting method of vector regression integrated network based on increment type | |
| CN112834927A (en) | Lithium battery residual life prediction method, system, device and medium | |
| WO2023035727A1 (en) | Industrial process soft-measurement method based on federated incremental stochastic configuration network | |
| CN107481523A (en) | A kind of traffic flow speed Forecasting Methodology and system | |
| CN108985455A (en) | A kind of computer application neural net prediction method and system | |
| CN107944550A (en) | A kind of space-time big data Forecasting Methodology based on detection type depth network | |
| CN104539601A (en) | Reliability analysis method and system for dynamic network attack process | |
| CN101834753A (en) | Network flow prediction method based on covariant orthogonality and neural network | |
| CN115982141A (en) | Characteristic optimization method for time series data prediction | |
| CN114611803A (en) | Switch device service life prediction method based on degradation characteristics | |
| CN112307536B (en) | Dam seepage parameter inversion method | |
| CN113627594A (en) | One-dimensional time sequence data amplification method based on WGAN | |
| CN111709519A (en) | Deep learning parallel computing architecture method and hyper-parameter automatic configuration optimization thereof | |
| CN115022195B (en) | Flow dynamic measurement method for IPv6 network | |
| CN110825583A (en) | Energy efficiency qualitative assessment technology for multi-index fusion of cloud data center |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211210 |